Martin Wendiggensen
2026
Cybermaxxing Models - Why Flexing Offensive Muscles Teaches Us How To Defend In The Age Of AI
AI is supposedly getting scarily good at cyber offense. Or so they tell us. But amid guardrails, trusted-access programs, hype cycles, and doomsday scenarios, the most capable models are increasingly gated when it comes to cybersecurity use cases. Unless you are actually at a frontier AI lab, most practitioners read the same public benchmarks and announcements. But who gets to use the models, and for what, is tightly controlled. This information scarcity poses a serious challenge. Â
If defense is indeed offense’s child, it is currently a child without a parent. Guardrails and export restrictions may keep us safer, but they also obscure our view of the offensive capabilities coming down the AI pipeline. Without a clear-eyed understanding of what is coming, adapting and preparing become much harder. This is especially relevant given that open-weights models lag behind but are still gaining ground. If today's gated capabilities are tomorrow’s public models, getting an early read is crucial. Â
Our talk will attempt to provide such an understanding. We push a combination of frontier and open-source models to a local maximum of their offensive capabilities. Along the way we discuss how to set up, measure and improve the use of models for offensive and defensive purposes. We demonstrate the things models are good at and speculate as to why and where they may break down. We demonstrate and discuss what learnings defenders can draw from continuous experimentation to engage and prepare for the new waves of capabilities.
About Martin Wendiggensen
Martin Wendiggensen is an AI Research Scientist at Dreadnode and PhD candidate at Johns Hopkins AIST. His research focuses on how AI is shifting the Cybersecurity Offensive-Defensive Balance. He received a Bachelor’s degree in Political Science with a focus on quantitative methods and Natural Language Processing from the University of Mannheim and studied in China, Israel, and Italy.Â
After working as policy advisor to a member of the German National Parliament’s Foreign Affairs Committee, Martin received a Master’s degree in International Relations from Johns Hopkins SAIS. He has conducted research at NATO as well as the University of Mannheim and specialized in leveraging High Performance Computing for applied research on AI capabilities at Hopkins.Â
Currently, he is a visiting researcher at ETH Zurich and writes a twice-monthly column explaining the technology of AI to a general audience for one of Europe’s largest newspapers.
